Module `Kaun.Loss`

Loss functions.

Losses are differentiable through Rune's autodiff and return scalar means. Invalid_argument messages are prefixed with Loss.<function>:.

Classification

val cross_entropy : (float, 'a) Nx.t -> (float, 'a) Nx.t -> (float, 'a) Nx.t

cross_entropy logits one_hot_labels is softmax cross-entropy.

logits has shape [...; num_classes] and must be rank >= 1. one_hot_labels must have the same shape.

Uses the log-sum-exp trick for numerical stability.

Raises Invalid_argument if ranks or shapes differ, or if num_classes is not positive.

val cross_entropy_sparse : 
  (float, 'a) Nx.t ->
  ('c, 'd) Nx.t ->
  (float, 'a) Nx.t

cross_entropy_sparse logits class_indices is cross_entropy with integer labels.

class_indices has shape [...] and must match logits without the last dimension. The class dimension is logits' last axis.

Raises Invalid_argument if labels are non-integer, ranks mismatch, non-class dimensions differ, or the class dimension is non-positive.

val binary_cross_entropy : 
  (float, 'a) Nx.t ->
  (float, 'a) Nx.t ->
  (float, 'a) Nx.t

binary_cross_entropy logits labels is sigmoid binary cross-entropy.

logits are raw (not sigmoid-normalized). labels are typically in [0;1]. Uses log-sigmoid for numerical stability.

Raises Invalid_argument if logits and labels shapes differ.

val mse : ('a, 'b) Nx.t -> ('a, 'b) Nx.t -> ('a, 'b) Nx.t

mse predictions targets is mean ((predictions - targets)^2).

Shape compatibility follows Nx broadcasting semantics.

val mae : ('a, 'b) Nx.t -> ('a, 'b) Nx.t -> ('a, 'b) Nx.t

mae predictions targets is mean (abs (predictions - targets)).

Shape compatibility follows Nx broadcasting semantics.